From 6792ec059696a358b6c98d8d86e9740b34c01e24 Mon Sep 17 00:00:00 2001 From: Florin Coras Date: Mon, 13 Mar 2017 03:49:51 -0700 Subject: TCP/session improvements - Added svm fifo flag for tracking fifo dequeue events (replaces event length). Updated all code to switch to the new scheme. - More session debugging - Fix peek index wrap - Add a trivial socket test client - Fast retransmit/cc fixes - tx and rx SACK fixes and unit testing - SRTT computation fix - remove dupack/ack burst filters - improve ack rx - improved segment rx - builtin client test code Change-Id: Ic4eb2d5ca446eb2260ccd3ccbcdaa73c64e7f4e1 Signed-off-by: Florin Coras Signed-off-by: Dave Barach --- src/vnet/tcp/tcp_input.c | 507 ++++++++++++++++++++++++++++++----------------- 1 file changed, 327 insertions(+), 180 deletions(-) (limited to 'src/vnet/tcp/tcp_input.c') diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c index 67af4321..5d11985f 100644 --- a/src/vnet/tcp/tcp_input.c +++ b/src/vnet/tcp/tcp_input.c @@ -95,13 +95,21 @@ vlib_node_registration_t tcp6_established_node; * or the rcv_nxt at last ack sent instead of rcv_nxt since that's the * peer's reference when computing our receive window. * - * This accepts only segments within the window. + * This: + * seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) && seq_geq (seq, tc->rcv_las) + * however, is too strict when we have retransmits. Instead we just check that + * the seq is not beyond the right edge and that the end of the segment is not + * less than the left edge. + * + * N.B. rcv_nxt and rcv_wnd are both updated in this node if acks are sent, so + * use rcv_nxt in the right edge window test instead of rcv_las. + * */ always_inline u8 tcp_segment_in_rcv_wnd (tcp_connection_t * tc, u32 seq, u32 end_seq) { - return seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) - && seq_geq (seq, tc->rcv_nxt); + return (seq_geq (end_seq, tc->rcv_las) + && seq_leq (seq, tc->rcv_nxt + tc->rcv_wnd)); } void @@ -253,6 +261,7 @@ tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0, { tcp_make_ack (tc0, b0); *next0 = tcp_next_output (tc0->c_is_ip4); + TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0); return -1; } } @@ -262,13 +271,25 @@ tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0, if (!tcp_segment_in_rcv_wnd (tc0, vnet_buffer (b0)->tcp.seq_number, vnet_buffer (b0)->tcp.seq_end)) { - if (!tcp_rst (th0)) + /* If our window is 0 and the packet is in sequence, let it pass + * through for ack processing. It should be dropped later.*/ + if (tc0->rcv_wnd == 0 + && tc0->rcv_nxt == vnet_buffer (b0)->tcp.seq_number) { - /* Send dup ack */ - tcp_make_ack (tc0, b0); - *next0 = tcp_next_output (tc0->c_is_ip4); + /* Make it look as if there's nothing to dequeue */ + vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number; + } + else + { + /* If not RST, send dup ack */ + if (!tcp_rst (th0)) + { + tcp_make_ack (tc0, b0); + *next0 = tcp_next_output (tc0->c_is_ip4); + TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0); + } + return -1; } - return -1; } /* 2nd: check the RST bit */ @@ -326,13 +347,13 @@ tcp_estimate_rtt (tcp_connection_t * tc, u32 mrtt) /* XXX Drop in RTT results in RTTVAR increase and bigger RTO. * The increase should be bound */ - tc->rttvar += (clib_abs (err) - tc->rttvar) >> 2; + tc->rttvar += ((int) clib_abs (err) - (int) tc->rttvar) >> 2; } else { /* First measurement. */ tc->srtt = mrtt; - tc->rttvar = mrtt << 1; + tc->rttvar = mrtt >> 1; } } @@ -394,7 +415,11 @@ tcp_dequeue_acked (tcp_connection_t * tc, u32 ack) } } -/** Check if dupack as per RFC5681 Sec. 2 */ +/** + * Check if dupack as per RFC5681 Sec. 2 + * + * This works only if called before updating snd_wnd. + * */ always_inline u8 tcp_ack_is_dupack (tcp_connection_t * tc, vlib_buffer_t * b, u32 new_snd_wnd) { @@ -429,10 +454,10 @@ scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) } sack_scoreboard_hole_t * -scoreboard_insert_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * prev, +scoreboard_insert_hole (sack_scoreboard_t * sb, u32 prev_index, u32 start, u32 end) { - sack_scoreboard_hole_t *hole, *next; + sack_scoreboard_hole_t *hole, *next, *prev; u32 hole_index; pool_get (sb->holes, hole); @@ -442,6 +467,7 @@ scoreboard_insert_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * prev, hole->end = end; hole_index = hole - sb->holes; + prev = scoreboard_get_hole (sb, prev_index); if (prev) { hole->prev = prev - sb->holes; @@ -462,28 +488,35 @@ scoreboard_insert_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * prev, return hole; } -static void +void tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) { sack_scoreboard_t *sb = &tc->sack_sb; sack_block_t *blk, tmp; - sack_scoreboard_hole_t *hole, *next_hole; - u32 blk_index = 0; + sack_scoreboard_hole_t *hole, *next_hole, *last_hole, *new_hole; + u32 blk_index = 0, old_sacked_bytes, hole_index; int i, j; - if (!tcp_opts_sack (tc) && sb->head == TCP_INVALID_SACK_HOLE_INDEX) + sb->last_sacked_bytes = 0; + sb->snd_una_adv = 0; + old_sacked_bytes = sb->sacked_bytes; + + if (!tcp_opts_sack (&tc->opt) && sb->head == TCP_INVALID_SACK_HOLE_INDEX) return; /* Remove invalid blocks */ - vec_foreach (blk, tc->opt.sacks) - { - if (seq_lt (blk->start, blk->end) - && seq_gt (blk->start, tc->snd_una) - && seq_gt (blk->start, ack) && seq_lt (blk->end, tc->snd_nxt)) - continue; - - vec_del1 (tc->opt.sacks, blk - tc->opt.sacks); - } + blk = tc->opt.sacks; + while (blk < vec_end (tc->opt.sacks)) + { + if (seq_lt (blk->start, blk->end) + && seq_gt (blk->start, tc->snd_una) + && seq_gt (blk->start, ack) && seq_leq (blk->end, tc->snd_nxt)) + { + blk++; + continue; + } + vec_del1 (tc->opt.sacks, blk - tc->opt.sacks); + } /* Add block for cumulative ack */ if (seq_gt (ack, tc->snd_una)) @@ -498,7 +531,7 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) /* Make sure blocks are ordered */ for (i = 0; i < vec_len (tc->opt.sacks); i++) - for (j = i; j < vec_len (tc->opt.sacks); j++) + for (j = i + 1; j < vec_len (tc->opt.sacks); j++) if (seq_lt (tc->opt.sacks[j].start, tc->opt.sacks[i].start)) { tmp = tc->opt.sacks[i]; @@ -506,10 +539,22 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) tc->opt.sacks[j] = tmp; } - /* If no holes, insert the first that covers all outstanding bytes */ if (sb->head == TCP_INVALID_SACK_HOLE_INDEX) { - scoreboard_insert_hole (sb, 0, tc->snd_una, tc->snd_una_max); + /* If no holes, insert the first that covers all outstanding bytes */ + last_hole = scoreboard_insert_hole (sb, TCP_INVALID_SACK_HOLE_INDEX, + tc->snd_una, tc->snd_una_max); + sb->tail = scoreboard_hole_index (sb, last_hole); + } + else + { + /* If we have holes but snd_una_max is beyond the last hole, update + * last hole end */ + tmp = tc->opt.sacks[vec_len (tc->opt.sacks) - 1]; + last_hole = scoreboard_last_hole (sb); + if (seq_gt (tc->snd_una_max, sb->max_byte_sacked) + && seq_gt (tc->snd_una_max, last_hole->end)) + last_hole->end = tc->snd_una_max; } /* Walk the holes with the SACK blocks */ @@ -526,10 +571,10 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) next_hole = scoreboard_next_hole (sb, hole); /* Byte accounting */ - if (seq_lt (hole->end, ack)) + if (seq_leq (hole->end, ack)) { - /* Bytes lost because snd wnd left edge advances */ - if (seq_lt (next_hole->start, ack)) + /* Bytes lost because snd_wnd left edge advances */ + if (next_hole && seq_leq (next_hole->start, ack)) sb->sacked_bytes -= next_hole->start - hole->end; else sb->sacked_bytes -= ack - hole->end; @@ -539,35 +584,78 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) sb->sacked_bytes += scoreboard_hole_bytes (hole); } + /* snd_una needs to be advanced */ + if (seq_geq (ack, hole->end)) + { + if (next_hole && seq_lt (ack, next_hole->start)) + sb->snd_una_adv = next_hole->start - ack; + else + sb->snd_una_adv = sb->max_byte_sacked - ack; + + /* all these can be delivered */ + sb->sacked_bytes -= sb->snd_una_adv; + } + + /* About to remove last hole */ + if (hole == last_hole) + { + sb->tail = hole->prev; + last_hole = scoreboard_last_hole (sb); + /* keep track of max byte sacked in case the last hole + * is acked */ + if (seq_gt (hole->end, sb->max_byte_sacked)) + sb->max_byte_sacked = hole->end; + } scoreboard_remove_hole (sb, hole); hole = next_hole; } - /* Partial overlap */ + /* Partial 'head' overlap */ else { - sb->sacked_bytes += blk->end - hole->start; - hole->start = blk->end; + if (seq_gt (blk->end, hole->start)) + { + sb->sacked_bytes += blk->end - hole->start; + hole->start = blk->end; + } blk_index++; } } else { /* Hole must be split */ - if (seq_leq (blk->end, hole->end)) + if (seq_lt (blk->end, hole->end)) { sb->sacked_bytes += blk->end - blk->start; - scoreboard_insert_hole (sb, hole, blk->end, hole->end); - hole->end = blk->start - 1; + hole_index = scoreboard_hole_index (sb, hole); + new_hole = scoreboard_insert_hole (sb, hole_index, blk->end, + hole->end); + + /* Pool might've moved */ + hole = scoreboard_get_hole (sb, hole_index); + hole->end = blk->start; + + /* New or split of tail */ + if ((last_hole->end == new_hole->end) + || seq_lt (last_hole->end, new_hole->start)) + { + last_hole = new_hole; + sb->tail = scoreboard_hole_index (sb, new_hole); + } + blk_index++; + hole = scoreboard_next_hole (sb, hole); } else { - sb->sacked_bytes += hole->end - blk->start + 1; - hole->end = blk->start - 1; + sb->sacked_bytes += hole->end - blk->start; + hole->end = blk->start; hole = scoreboard_next_hole (sb, hole); } } } + + sb->last_sacked_bytes = sb->sacked_bytes + sb->snd_una_adv + - old_sacked_bytes; } /** Update snd_wnd @@ -577,72 +665,94 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) static void tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd) { - if (tc->snd_wl1 < seq || (tc->snd_wl1 == seq && tc->snd_wl2 <= ack)) + if (seq_lt (tc->snd_wl1, seq) + || (tc->snd_wl1 == seq && seq_leq (tc->snd_wl2, ack))) { tc->snd_wnd = snd_wnd; tc->snd_wl1 = seq; tc->snd_wl2 = ack; + TCP_EVT_DBG (TCP_EVT_SND_WND, tc); } } -static void +void tcp_cc_congestion (tcp_connection_t * tc) { + tc->snd_congestion = tc->snd_nxt; tc->cc_algo->congestion (tc); + TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4); } -static void +void tcp_cc_recover (tcp_connection_t * tc) { - if (tcp_in_fastrecovery (tc)) - { - tc->cc_algo->recovered (tc); - tcp_recovery_off (tc); - } - else if (tcp_in_recovery (tc)) - { - tcp_recovery_off (tc); - tc->cwnd = tcp_loss_wnd (tc); - } + tc->cc_algo->recovered (tc); + + tc->rtx_bytes = 0; + tc->rcv_dupacks = 0; + tc->snd_nxt = tc->snd_una; + + tc->cc_algo->rcv_ack (tc); + tc->tsecr_last_ack = tc->opt.tsecr; + + tcp_fastrecovery_1_smss_off (tc); + tcp_fastrecovery_off (tc); + + TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3); } static void -tcp_cc_rcv_ack (tcp_connection_t * tc) +tcp_cc_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b) { u8 partial_ack; if (tcp_in_recovery (tc)) { - partial_ack = seq_lt (tc->snd_una, tc->snd_una_max); + partial_ack = seq_lt (tc->snd_una, tc->snd_congestion); if (!partial_ack) { /* Clear retransmitted bytes. */ - tc->rtx_bytes = 0; tcp_cc_recover (tc); } else { + TCP_EVT_DBG (TCP_EVT_CC_PACK, tc); + /* Clear retransmitted bytes. XXX should we clear all? */ tc->rtx_bytes = 0; tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK); - /* Retransmit first unacked segment */ - tcp_retransmit_first_unacked (tc); + /* In case snd_nxt is still in the past and output tries to + * shove some new bytes */ + tc->snd_nxt = tc->snd_una; + + /* XXX need proper RFC6675 support */ + if (tc->sack_sb.last_sacked_bytes) + { + tcp_fast_retransmit (tc); + } + else + { + /* Retransmit first unacked segment */ + tcp_retransmit_first_unacked (tc); + /* If window allows, send 1 SMSS of new data */ + if (seq_lt (tc->snd_nxt, tc->snd_congestion)) + tc->snd_nxt = tc->snd_congestion; + } } } else { tc->cc_algo->rcv_ack (tc); + tc->tsecr_last_ack = tc->opt.tsecr; + tc->rcv_dupacks = 0; } - - tc->rcv_dupacks = 0; - tc->tsecr_last_ack = tc->opt.tsecr; } static void tcp_cc_rcv_dupack (tcp_connection_t * tc, u32 ack) { - ASSERT (tc->snd_una == ack); +// ASSERT (seq_geq(tc->snd_una, ack)); tc->rcv_dupacks++; if (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) @@ -688,20 +798,39 @@ tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, { u32 new_snd_wnd; - /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) then send an - * ACK, drop the segment, and return */ + /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) */ if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)) { - tcp_make_ack (tc, b); - *next = tcp_next_output (tc->c_is_ip4); - *error = TCP_ERROR_ACK_INVALID; - return -1; + /* If we have outstanding data and this is within the window, accept it, + * probably retransmit has timed out. Otherwise ACK segment and then + * drop it */ + if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_una_max)) + { + tcp_make_ack (tc, b); + *next = tcp_next_output (tc->c_is_ip4); + *error = TCP_ERROR_ACK_INVALID; + TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 0, + vnet_buffer (b)->tcp.ack_number); + return -1; + } + + tc->snd_nxt = vnet_buffer (b)->tcp.ack_number; + *error = TCP_ERROR_ACK_FUTURE; + TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 2, + vnet_buffer (b)->tcp.ack_number); } - /* If old ACK, discard */ + /* If old ACK, probably it's an old dupack */ if (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)) { *error = TCP_ERROR_ACK_OLD; + TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1, + vnet_buffer (b)->tcp.ack_number); + if (tcp_in_fastrecovery (tc) && tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) + { + TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc); + tcp_cc_rcv_dupack (tc, vnet_buffer (b)->tcp.ack_number); + } return -1; } @@ -712,32 +841,40 @@ tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, if (tcp_ack_is_dupack (tc, b, new_snd_wnd)) { + TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1); tcp_cc_rcv_dupack (tc, vnet_buffer (b)->tcp.ack_number); *error = TCP_ERROR_ACK_DUP; return -1; } - /* Valid ACK */ + /* + * Valid ACK + */ + tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una; - tc->snd_una = vnet_buffer (b)->tcp.ack_number; + tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv; - /* Dequeue ACKed packet and update RTT */ + /* Dequeue ACKed data and update RTT */ tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number); - tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number, vnet_buffer (b)->tcp.ack_number, new_snd_wnd); - /* Updates congestion control (slow start/congestion avoidance) */ - tcp_cc_rcv_ack (tc); + /* If some of our sent bytes have been acked, update cc and retransmit + * timer. */ + if (tc->bytes_acked) + { + TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc, vnet_buffer (b)->tcp.ack_number); - TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc); + /* Updates congestion control (slow start/congestion avoidance) */ + tcp_cc_rcv_ack (tc, b); - /* If everything has been acked, stop retransmit timer - * otherwise update */ - if (tc->snd_una == tc->snd_una_max) - tcp_timer_reset (tc, TCP_TIMER_RETRANSMIT); - else - tcp_timer_update (tc, TCP_TIMER_RETRANSMIT, tc->rto); + /* If everything has been acked, stop retransmit timer + * otherwise update */ + if (tc->snd_una == tc->snd_una_max) + tcp_retransmit_timer_reset (tc); + else + tcp_retransmit_timer_update (tc); + } return 0; } @@ -757,9 +894,7 @@ static void tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end) { sack_block_t *new_list = 0, block; - u32 n_elts; int i; - u8 new_head = 0; /* If the first segment is ooo add it to the list. Last write might've moved * rcv_nxt over the first segment. */ @@ -768,7 +903,6 @@ tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end) block.start = start; block.end = end; vec_add1 (new_list, block); - new_head = 1; } /* Find the blocks still worth keeping. */ @@ -782,20 +916,19 @@ tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end) || seq_leq (tc->snd_sacks[i].start, end)) continue; - /* Save subsequent segments to new SACK list. */ - n_elts = clib_min (vec_len (tc->snd_sacks) - i, - TCP_MAX_SACK_BLOCKS - new_head); - vec_insert_elts (new_list, &tc->snd_sacks[i], n_elts, new_head); - break; + /* Save to new SACK list. */ + vec_add1 (new_list, tc->snd_sacks[i]); } + ASSERT (vec_len (new_list) < TCP_MAX_SACK_BLOCKS); + /* Replace old vector with new one */ vec_free (tc->snd_sacks); tc->snd_sacks = new_list; } /** Enqueue data for delivery to application */ -always_inline u32 +always_inline int tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, u16 data_len) { @@ -812,6 +945,8 @@ tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, vlib_buffer_get_current (b), data_len, 1 /* queue event */ ); + TCP_EVT_DBG (TCP_EVT_INPUT, tc, 0, data_len, written); + /* Update rcv_nxt */ if (PREDICT_TRUE (written == data_len)) { @@ -824,38 +959,61 @@ tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, /* Send ACK confirming the update */ tc->flags |= TCP_CONN_SNDACK; + } + else if (written > 0) + { + /* We've written something but FIFO is probably full now */ + tc->rcv_nxt += written; - /* Update SACK list if need be */ - if (tcp_opts_sack_permitted (&tc->opt)) - { - /* Remove SACK blocks that have been delivered */ - tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt); - } + /* Depending on how fast the app is, all remaining buffers in burst will + * not be enqueued. Should we inform peer of the damage? XXX */ + return TCP_ERROR_PARTIALLY_ENQUEUED; } else { - ASSERT (0); return TCP_ERROR_FIFO_FULL; } + /* Update SACK list if need be */ + if (tcp_opts_sack_permitted (&tc->opt)) + { + /* Remove SACK blocks that have been delivered */ + tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt); + } + return TCP_ERROR_ENQUEUED; } /** Enqueue out-of-order data */ -always_inline u32 +always_inline int tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b, u16 data_len) { stream_session_t *s0; u32 offset, seq; + int rv; + + /* Pure ACK. Do nothing */ + if (PREDICT_FALSE (data_len == 0)) + { + return TCP_ERROR_PURE_ACK; + } s0 = stream_session_get (tc->c_s_index, tc->c_thread_index); seq = vnet_buffer (b)->tcp.seq_number; offset = seq - tc->rcv_nxt; - if (svm_fifo_enqueue_with_offset (s0->server_rx_fifo, s0->pid, offset, - data_len, vlib_buffer_get_current (b))) - return TCP_ERROR_FIFO_FULL; + rv = svm_fifo_enqueue_with_offset (s0->server_rx_fifo, s0->pid, offset, + data_len, vlib_buffer_get_current (b)); + + /* Nothing written */ + if (rv) + { + TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, 0); + return TCP_ERROR_FIFO_FULL; + } + + TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len); /* Update SACK list if in use */ if (tcp_opts_sack_permitted (&tc->opt)) @@ -875,20 +1033,23 @@ tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b, } /** - * Check if ACK could be delayed. DELACK timer is set only after frame is - * processed so this can return true for a full bursts of packets. + * Check if ACK could be delayed. If ack can be delayed, it should return + * true for a full frame. If we're always acking return 0. */ always_inline int tcp_can_delack (tcp_connection_t * tc) { - /* If there's no DELACK timer set and the last window sent wasn't 0 we - * can safely delay. */ - if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK) - && (tc->flags & TCP_CONN_SENT_RCV_WND0) == 0 - && (tc->flags & TCP_CONN_SNDACK) == 0) - return 1; + /* Send ack if ... */ + if (TCP_ALWAYS_ACK + /* just sent a rcv wnd 0 */ + || (tc->flags & TCP_CONN_SENT_RCV_WND0) != 0 + /* constrained to send ack */ + || (tc->flags & TCP_CONN_SNDACK) != 0 + /* we're almost out of tx wnd */ + || tcp_available_snd_space (tc) < 2 * tc->snd_mss) + return 0; - return 0; + return 1; } static int @@ -900,23 +1061,33 @@ tcp_segment_rcv (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b, /* Handle out-of-order data */ if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt)) { + /* Old sequence numbers allowed through because they overlapped + * the rx window */ + if (seq_lt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt)) + { + error = TCP_ERROR_SEGMENT_OLD; + *next0 = TCP_NEXT_DROP; + goto done; + } + error = tcp_session_enqueue_ooo (tc, b, n_data_bytes); - /* Don't send more than 3 dupacks per burst - * XXX decide if this is good */ - if (tc->snt_dupacks < 3) - { - /* RFC2581: Send DUPACK for fast retransmit */ - tcp_make_ack (tc, b); - *next0 = tcp_next_output (tc->c_is_ip4); + /* N.B. Should not filter burst of dupacks. Two issues 1) dupacks open + * cwnd on remote peer when congested 2) acks leaving should have the + * latest rcv_wnd since the burst may eaten up all of it, so only the + * old ones could be filtered. + */ - /* Mark as DUPACK. We may filter these in output if - * the burst fills the holes. */ - vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK; + /* RFC2581: Send DUPACK for fast retransmit */ + tcp_make_ack (tc, b); + *next0 = tcp_next_output (tc->c_is_ip4); - tc->snt_dupacks++; - } + /* Mark as DUPACK. We may filter these in output if + * the burst fills the holes. */ + if (n_data_bytes) + vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK; + TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc); goto done; } @@ -924,63 +1095,45 @@ tcp_segment_rcv (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b, * segments can be enqueued after fifo tail offset changes. */ error = tcp_session_enqueue_data (tc, b, n_data_bytes); - TCP_EVT_DBG (TCP_EVT_INPUT, tc, n_data_bytes); + if (n_data_bytes == 0) + { + *next0 = TCP_NEXT_DROP; + goto done; + } + + if (PREDICT_FALSE (error == TCP_ERROR_FIFO_FULL)) + *next0 = TCP_NEXT_DROP; /* Check if ACK can be delayed */ - if (tcp_can_delack (tc)) + if (!tcp_can_delack (tc)) { - /* Nothing to do for pure ACKs */ + /* Nothing to do for pure ACKs XXX */ if (n_data_bytes == 0) goto done; - /* If connection has not been previously marked for delay ack - * add it to the list and flag it */ - if (!tc->flags & TCP_CONN_DELACK) - { - vec_add1 (tm->delack_connections[tc->c_thread_index], - tc->c_c_index); - tc->flags |= TCP_CONN_DELACK; - } + *next0 = tcp_next_output (tc->c_is_ip4); + tcp_make_ack (tc, b); } else { - /* Check if a packet has already been enqueued to output for burst. - * If yes, then drop this one, otherwise, let it pass through to - * output */ - if ((tc->flags & TCP_CONN_BURSTACK) == 0) - { - *next0 = tcp_next_output (tc->c_is_ip4); - tcp_make_ack (tc, b); - error = TCP_ERROR_ENQUEUED; - - /* TODO: maybe add counter to ensure N acks will be sent/burst */ - tc->flags |= TCP_CONN_BURSTACK; - } + if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK)) + tcp_timer_set (tc, TCP_TIMER_DELACK, TCP_DELACK_TIME); } done: return error; } -void -delack_timers_init (tcp_main_t * tm, u32 thread_index) +always_inline void +tcp_established_inc_counter (vlib_main_t * vm, u8 is_ip4, u8 evt, u8 val) { - tcp_connection_t *tc; - u32 i, *conns; - tw_timer_wheel_16t_2w_512sl_t *tw; - - tw = &tm->timer_wheels[thread_index]; - conns = tm->delack_connections[thread_index]; - for (i = 0; i < vec_len (conns); i++) - { - tc = pool_elt_at_index (tm->connections[thread_index], conns[i]); - ASSERT (0 != tc); + if (PREDICT_TRUE (!val)) + return; - tc->timers[TCP_TIMER_DELACK] - = tw_timer_start_16t_2w_512sl (tw, conns[i], - TCP_TIMER_DELACK, TCP_DELACK_TIME); - } - vec_reset_length (tm->delack_connections[thread_index]); + if (is_ip4) + vlib_node_increment_counter (vm, tcp4_established_node.index, evt, val); + else + vlib_node_increment_counter (vm, tcp6_established_node.index, evt, val); } always_inline uword @@ -1027,7 +1180,7 @@ tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, if (PREDICT_FALSE (tc0 == 0)) { error0 = TCP_ERROR_INVALID_CONNECTION; - goto drop; + goto done; } /* Checksum computed by ipx_local no need to compute again */ @@ -1061,18 +1214,22 @@ tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, th0, &next0))) { error0 = TCP_ERROR_SEGMENT_INVALID; - goto drop; + TCP_EVT_DBG (TCP_EVT_SEG_INVALID, tc0, + vnet_buffer (b0)->tcp.seq_number, + vnet_buffer (b0)->tcp.seq_end); + goto done; } /* 5: check the ACK field */ if (tcp_rcv_ack (tc0, b0, th0, &next0, &error0)) { - goto drop; + goto done; } /* 6: check the URG bit TODO */ /* 7: process the segment text */ + vlib_buffer_advance (b0, n_advance_bytes0); error0 = tcp_segment_rcv (tm, tc0, b0, n_data_bytes0, &next0); @@ -1088,7 +1245,7 @@ tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME); } - drop: + done: b0->error = node->errors[error0]; if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { @@ -1103,17 +1260,7 @@ tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, } errors = session_manager_flush_enqueue_events (my_thread_index); - if (errors) - { - if (is_ip4) - vlib_node_increment_counter (vm, tcp4_established_node.index, - TCP_ERROR_EVENT_FIFO_FULL, errors); - else - vlib_node_increment_counter (vm, tcp6_established_node.index, - TCP_ERROR_EVENT_FIFO_FULL, errors); - } - - delack_timers_init (tm, my_thread_index); + tcp_established_inc_counter (vm, is_ip4, TCP_ERROR_EVENT_FIFO_FULL, errors); return from_frame->n_vectors; } @@ -1602,7 +1749,7 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, stream_session_accept_notify (&tc0->connection); /* Reset SYN-ACK retransmit timer */ - tcp_timer_reset (tc0, TCP_TIMER_RETRANSMIT); + tcp_retransmit_timer_reset (tc0); break; case TCP_STATE_ESTABLISHED: /* We can get packets in established state here because they @@ -1668,7 +1815,7 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME); /* Stop retransmit */ - tcp_timer_reset (tc0, TCP_TIMER_RETRANSMIT); + tcp_retransmit_timer_reset (tc0); goto drop; -- cgit 1.2.3-korg